Asus designs most of its high-end motherboards with a wide variety of overclocking, voltage, and timing adjustments beyond what's needed to push most hardware to its limit. This would be lab technician Shelton Romhanyi's chance to greatly expand his knowledge of the art.
It's easy for an experienced overclocker to forget how steep the learning curve can be for first timers. Shelton encountered a heat monster and spent quite some time tracking down the cause to poor case ventilation. Only after completely benchmarking the first overclock attempt did he contact anyone for advice, and a 12-hour Q&A session ensued.
Rather than throw out the initial results, we kept them to show what the original configuration could do.
|Overclocked Component Settings|
|Processor||Intel Core 2 Extreme X6800, 2.93 GHz, 1066 FSB, 4 MB Cache|
|First Overclock||3.025 GHz, 11x 275 MHz FSB, 1.60 Volts Core|
|Final Overclock||3.663 GHz: 11x 333 MHz FSB, 1.55 Volts Core|
|Motherboard||ASUS Striker Extreme, BIOS: Phoenix 0901 (02/07/07)|
|First Overclock||PCI-E SPP & MCP: 3450 MHz @ 1.45V|
|Final Overclock||PCI-E SPP & MCP: 3450 MHz @ 1.45V|
|RAM||Crucial Ballistix PC2-8000, 2x 1024 MB, CAS 5.0-5-5-15|
|First Overclock||1100 MHz Data Rate, CAS 4.0-4-4-6, 1.85 Volts|
|Final Overclock||950 MHz Data Rate, CAS 4.0-4-4-6, 2.20 Volts|
|Graphics Cards||2x XFX GeForce 8800 GTX, 630 MHz GPU, 1800 MHz RAM|
Normally we'd recommend using 1.50 volts at the 680i Northbridge to assist chipset overclocking capabilities, but this much voltage wasn't needed because we reached the second heat limit of the CPU before reaching the chipset's stability limit. Note that the 680i runs hot, so any reduction in voltage could potentially pay off in a heat reduction, leading to a more stable CPU.
As we increased CPU FSB, the memory became unstable even as we tried to keep its speed relatively constant. Shelton thought that the RAM might have been damaged, but reducing its data rate and increasing its voltage got it stable again. Notice that we made every effort to keep latencies low, as reduced latencies seem to benefit program performance far better than memory clock rates in excess of 1.5 times the FSB clock. Remember that this system is in dual-channel mode, which compensates for the CPU FSB being quad data rate when tied to double data rate memory.
Using a case that provides even greater airflow past the CPU could further improve these results, as could using a later CPU core stepping than our pre-retail Intel Core 2 Extreme X6800 sample. Such changes would do nothing to validate a system we've carried forward from our previous System Builder Marathon, however.
As a final note, Shelton did not overclock the graphics cards. Did he forget, or were they already too close to their limits? Asking such questions would get us nowhere, as there's simply no time left to retest.